Siebel Smart Answer Administration Guide > Configuration and Implementation of Siebel Smart Answer >

Creating and Training Your KB

Following successful installation of Banter Workbench tool, and creation and categorization of your corpus file, you are now ready to create and train your KB.

You should perform the following tasks in the sequence shown:

  1. Create and analyze your KB in Workbench.

    2. If you want to create a hierarchical KB, use the KB Editor to define the structure and use this KB to apply training and analysis on.

    When you use Workbench to create and analyze in the same process, the main corpus is divided into two parts. One corpus is used to create the KB (known as the training corpus) and the other (the testing corpus) is used to analyze performance of the KB when classifying data.

  2. Train and test the KB within Workbench.

    3. In general, increasing the number of items you use for training the KB improves its performance. Therefore, before you introduce the KB into a "live" production environment, use the Create, Analyze and Learn wizard and create using "Create using all selected, analyze using all selected" option just as a final check before introducing your KB into a production environment.

    After you begin using a KB file to categorize incoming email, and requests and users select categories or apply feedback, the feedback is stored in the KB file on the Siebel File system.

    CAUTION:  After you begin receiving feedback on a KB file, you should never train that same KB file again. If you do, you will delete all the statistical learning information based on your feedback.

  3. Analyze training results by reviewing Workbench graphs and reports.

    4. The Cumulative Success Graph provides an overall view of KB performance. The Precision vs. Recall graphs help to determine category thresholds for generating automatic responses. For more information about thresholds, see Modifying a Category.


 Siebel Smart Answer Administration Guide 
 Published: 23 June 2003